LANDSLIDE SUSCEPTIBILITY ZONATION OF RONGGA AND SURROUNDING AREAS USING WEIGHT OF EVIDENCE (WOE), LOGISTIC REGRESSION (LR) AND COMBINATION WOE-LR METHOD

• Main Author: Bahtera Prasaja, Dika
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/53480
• Institution: Institut Teknologi Bandung

Indonesia is a country with a very complex natural disaster condition. One of the natural disasters that often occurs is landslide. Landslide is the movement of rock and or soil masses which is influenced by controlling factors and triggering factors. Controlling factors include the presence of steep to very steep slope morphology, presence of geological structures, and less resistant rock conditions. In general, the triggering factors are the conditions of rainfall and hidrology in an area. Research on disaster conditions in an area is needed to prevent and reduce negative impacts. The research area located in Rongga District, West Bandung Regency, which has a morphology of steep hills that make the area prone to landslide. On March 23, 2020, there was a landslide on a 30 m high cliff in Nyomplong Village, Cibitung Village. The incident was triggered by heavy rain and intense. Although there were no casualties, 37 people had to be evacuated due to the proximity of the buildings to the scene. The research was conducted by analyzing landslide modeling using bivariate statistics: Weight of Evidence (WoE) Method, multivariate: Logistic Regression (LR) Method and combination of the WoE-LR Method. The data analyzed were the parameters of landslide and the location of landslide. It is recorded that in the research area there are 572 locations of landslides taken based on field data and appearances from Google Earth. The data is divided into two groups: analysis test data (ls train) with a percentage of 70% and validation test data (ls test) with a percentage of 30%. The parameters used in the analysis of landslide are: land use, slope, slope direction, curvature, elevation, rainfall, lithology, NDWI, NDVI, distance from the road, distance from the river, flow direction, lineaments density, river density, and distance from lineaments. This parameter is tested for validation by determining the value of the area under curve (AUC). There are 10 parameters that passed the AUC test (AUC> 0.6): land use (0.60), slope (0.68), curvature (0.61), elevation (0.63), rainfall (0.60), lithology (0.65), river density (0.62), geological structure and lineament density (0,61), NDVI (0,61) and distance from the river (0.60). Validation map zonation for each method is to determine the value of AUC (area under curve), SCAI (seed cell area index), and spatial domain. The AUC value calculated is the AUC success rate and AUC prediction rate. The success rate is obtained by combining the sum of WoE data with the landslide test data (ls_train), totaling 400 landslide locations. The prediction rate is obtained by combining the total WoE data with the landslide test data (ls_test), totaling 172 points of landslide locations. Success rate AUC value is 0.69 and prediction rate AUC value is 0.66 for the WoE Method. Success rate AUC value is 0.70 and prediction rate AUC value is 0.69 LR Method. Success rate AUC value is 0.71 and prediction rate AUC value is 0.66 for the WoE-LR Method. SCAI value results show that are not too far from each method. WoE-LR Method maps have better results for very low to low landslide susceptibility zonation with the greatest SCAI value and WoE Method maps have better results for high landslide susceptibility zonation with the highest SCAI value. small. The results of validation using the spatial domain show that all methods have correct and acceptable pixels above 90% of the total area. The largest spatial domain value is obtained from the validation of the LR and WoE-LR Methods: percentage of pixels that are classified correctly is 56.2%, acceptable pixels are 42.9%, and unacceptable pixels are less than 1%. Pixel values that are correctly classified and can be accepted indicate the accuracy level of the modelling.